Last updated: October 2026
Bad audio rarely ruins a vlog in one way. It hisses, it rumbles, it echoes off the walls, and each fault needs a different tool in a different order. This guide works symptom by symptom: play your clip, name what you hear, then apply the matching fix with tools already in most editors. It ends with YouTube’s loudness target, and the faults no editor can save.
The short version
- Diagnose first. Listen on headphones, and listen to the pauses. The gaps between sentences name the fault.
- Steady noise yields to a noise profile. Sample it alone, reduce gently; two light passes beat one heavy one.
- Wind and rumble are low frequency problems. A high-pass filter near 80 to 100 Hz does most of the work.
- Order matters. Trim and level, high-pass, noise reduction, EQ, de-reverb, then limiter and loudness last.
- Clipping is the exception. Distortion baked into loud words cannot be fully undone. Some takes must be recorded again.
On this page
- Name the fault before you touch a slider
- Steady noise: fans, air con and the hiss under everything
- Wind rumble and handling thumps
- Mains hum and electrical buzz
- Echo and reverb: the recorded-in-a-bathroom voice
- Clipping and distortion: the fault with no undo
- EQ moves that rescue a boxy, muddy or harsh voice
- The repair chain: fix things in this order
- Loudness: the YouTube target most guides skip
- When the honest fix is to record it again
- Bad audio questions, answered
- Keep exploring
Name the fault before you touch a slider
Put headphones on, turn the clip up, and listen to the gaps between sentences. Whatever fills them is your fault: a steady wash means constant noise, thumps mean wind, a constant tone means hum, a smeared repeat means reverb, crackle on loud words means clipping.
Most failed cleanups start with the wrong tool. Noise reduction aimed at echo does nothing except chew the voice, and EQ aimed at wind just reshapes the rumble. Solo the track if your editor allows it, loop ten seconds including a pause, and match what you hear to the table.
| What you hear | What it actually is | First tool to reach for |
|---|---|---|
| A constant hiss, roar or whir under everything | Steady noise: fan, air con, fridge, preamp hiss | Noise reduction with a sampled profile |
| Low thumps and roaring that swells and drops | Wind on the mic, or hands moving on the camera | High-pass filter around 80 to 100 Hz |
| A constant low tone or buzzing | Mains hum from power supplies, chargers or lighting | Narrow notch EQ at the hum frequency |
| Voice sounds distant, a smeared copy trailing it | Room echo and reverb from hard surfaces | De-reverb, helped by a gate or expander |
| Harsh crackle on the loudest words | Clipping: the level ran past the maximum | De-clip tool if mild, re-record if heavy |
Real clips often carry two faults at once, typically echo over fan noise. Fix one at a time, in the repair chain order below, and re-listen after each step. The waveform tells the same story: healthy recordings drop to a thin line between sentences; noisy ones keep a thick fuzzy band.
Steady noise: fans, air con and the hiss under everything
Constant noise is the easiest fault to fix because it never changes. Find a second or two where only the noise plays, teach it to the tool as a noise profile, then apply a gentle reduction to the whole clip. Tame the noise; do not try to erase it.
The profile method works because the tool can only subtract what it can recognise, so select a stretch with no speech, breaths or clicks, just the fan or hiss alone. That sample is the Get Noise Profile step in Audacity, the noise reduction in Premiere Pro’s Essential Sound panel, and the denoise in most of the free editing software for vlogging in our roundup.
Restraint decides whether the result sounds clean or strange. Push reduction hard and the voice turns watery and metallic, because the tool eats speech frequencies overlapping the noise. Start modestly, roughly 6 to 10 dB, judge the voice rather than the silence, and add a second light pass if needed.
AI voice isolation, the one-slider cleanup in several editors, works the opposite way: it learns what a voice sounds like and rebuilds the track around it. It is fast and often startlingly good on fan noise. Keep strength part-way, and check s and f sounds afterwards.
Wind rumble and handling thumps
Wind noise is mostly very low frequency energy, big slow pressure waves that swamp the mic. A high-pass filter near 80 to 100 Hz cuts it away while leaving the voice, which lives higher up, almost untouched. It is the most effective wind fix in the edit.
Noise profiles fail on wind because a profile assumes the noise stays the same, while wind changes character every second. The filter ignores pattern and only cares about pitch. Drag the high-pass corner up slowly while a gust plays: the roar falls away first, and the voice thins if you go too far. Most voices sit best between 80 and 120 Hz.
What the filter cannot save is a gust that distorts, or a hand thump mid-sentence. Those spots are short: cut them if the sentence survives, or dip the volume for that half second. A vlog with two muted gusts plays fine; one where every third sentence vanishes belongs in the re-record section.
Mains hum and electrical buzz
Hum is a pure tone leaking in from mains power: 60 Hz in the Americas, 50 Hz in most of the rest of the world, plus fainter copies at double and triple that frequency. A narrow notch EQ cut at each line removes it cleanly, because almost nothing in a voice sits exactly on those frequencies.
Hum gets lumped in with hiss and attacked with noise reduction, which smears it instead of removing it. The notch method is surgical: create a narrow EQ band with a high Q value, park it on 50 or 60 Hz depending on your country, and cut 10 dB or more. The tone should vanish from the gaps while the voice stays put. Repeat at the harmonics, 100 or 120 Hz, then 150 or 180 Hz.
Buzz is hum’s rougher cousin, usually from dimmer switches, cheap LED lighting or a phone charger sharing the circuit. It spreads across more harmonics, so expect three or four notches. If it changes when a charger is unplugged, you have found the source.
Echo and reverb: the recorded-in-a-bathroom voice
Echo is your own voice reaching the mic twice, directly and again bounced off walls, floor and ceiling. De-reverb tools estimate the bounced copy and subtract part of it, pulling the voice closer and drier. They cannot rebuild a voice recorded across a bare room.
Set expectations first, because cleanup guides overpromise here. The direct voice and its reflections overlap in time and frequency, so tools separate them by educated guessing. Mild room sound cleans up well; a tiled kitchen recorded from three metres away improves but never sounds like a studio.
Work in this order. Run de-reverb at a moderate setting and judge it on headphones. Then try a gate or expander with a gentle ratio: it lowers the volume in the gaps between phrases, where reverb tails hang exposed. Keep the threshold low enough that quiet word endings survive. A small EQ cut around 300 to 500 Hz then trims boxiness.
Distance is the real cause: the further the mic sat from the mouth, the more bounced sound the recording holds, which is why the same room sounds fine on a close lav and terrible on a camera mic across the room. Our guide to microphone patterns and placement explains that distance problem.
Clipping and distortion: the fault with no undo
Clipping happens when the recording level runs past the loudest signal the recorder can store. The waveform tops are sliced flat and the detail in those moments is not in the file. De-clip tools guess at mild cases; heavy clipping is permanent.
You can see clipping before you hear it: the waveform looks ironed against the ceiling, flat-topped on loud words. Mild clipping, a few peaks on the loudest shouts, often responds to a de-clip tool, which rebuilds rounded peaks from the surrounding shape. When every other word crackles, the result stays harsh.
The mirror problem is a recording made far too quiet, then turned up. Gain raises the noise floor by the same amount, so the hiss climbs with the voice and their ratio never improves. Denoise at the recorded level first, while the noise is easy to profile, and raise level later, at the loudness stage.
Neither problem is really an editing problem. Both are level problems set at the moment of recording, which is why the re-record section treats repeated clipping as a stop-and-reshoot fault rather than a challenge for more plugins.
EQ moves that rescue a boxy, muddy or harsh voice
Most voice problems live in predictable places. Rumble sits below 80 Hz, mud and boxiness pile up between 250 and 500 Hz, clarity lives around 3 to 5 kHz, harshness bites between 6 and 9 kHz. Small cuts of 1 to 3 dB in the problem zone beat big boosts.
Start with the high-pass filter from the wind section: for indoor talking-head clips, rolling off below 75 to 85 Hz removes desk thumps and handling noise that muddies phone speakers. Then listen for boxiness, the talking-into-a-cupboard quality small rooms add. A wide, gentle cut around 300 to 450 Hz, just 2 dB, opens the voice up more than any other move.
If the voice still sounds dull, add a small broad lift of 1 to 3 dB around 3 to 5 kHz, the presence region where consonants live. A lift there reads as clarity, not loudness. For harshness, reverse it: boost one narrow band, sweep it across the top range over a bright passage, and cut the spot that makes you wince, by 2 to 4 dB.
Keep every move small and check against the bypass switch. EQ is easy to overdo because your ears adapt within seconds and keep asking for more. If bypassing makes the voice sound worse in a way you can name, the move is right; if you hear no difference, it was decoration.
The repair chain: fix things in this order
Work in six steps: trim and level, high-pass filter, noise reduction, EQ, de-reverb, then limiter and loudness. Each step hands the next a cleaner signal; out of sequence, the tools fight each other.
- Trim and level. Cut mistakes and dead air first, and bring clips to the same rough level. Repairing audio you are about to delete wastes time, and level jumps make later tools behave differently per clip.
- High-pass filter. Remove rumble and wind energy before noise reduction sees the track. Low frequency energy is loud, and it confuses profiling tools into treating the track as bass heavy.
- Noise reduction. Profile and reduce the steady noise on the cleaned-up track. This step decides the noise floor for everything downstream.
- EQ. Shape the voice after the noise is down. Boosting presence first also boosts the hiss in that region, forcing the reduction to run again.
- De-reverb. Dry the voice out once its tone is settled. De-reverb tools judge reflections partly by tone, so they behave more predictably after EQ.
- Limiter and loudness. Set final levels last. A limiter placed earlier pumps the noise floor up in every pause and fights every later change.
Save the chain as a preset once it works: vloggers record in the same rooms with the same mics, so last week’s chain usually starts this week’s video. It is one of the workflow shortcuts in our guide to how to edit vlogs faster.
Loudness: the YouTube target most guides skip
YouTube normalises playback to about -14 LUFS integrated, with true peaks at or below -1 dBTP. Measure the finished video as a whole, use the limiter last to hold peaks under the ceiling, and upload at that target instead of as loud as possible.
LUFS measures the average loudness of the whole video as a listener experiences it, which beats peak meters: two vlogs can peak identically yet feel completely different if one voice sits 6 dB hotter throughout. Most editors include a loudness meter or normalise-to-LUFS export; free tools can measure a finished file.
Upload much louder than the target and YouTube turns your video down, so the extra loudness bought nothing but a squashed sound. Upload much quieter and it largely stays quieter, because raising you would push your peaks through the ceiling. A quiet vlog beside everyone else’s normalised videos feels weak. Aim for -14 LUFS integrated, true peaks at -1 dBTP or lower, and a consistent voice throughout.
Clean, consistent dialogue also pays off in captions: automatic captions are generated from your audio, and steady levels with a low noise floor give the recogniser less to get wrong. Our guide to captions and subtitles for vlogs covers that workflow in full.
When the honest fix is to record it again
Re-record when the fault is louder than the fix: clipping through most sentences, wind swamping whole passages, noise close under the voice all clip long, or echo so heavy de-reverb leaves words unintelligible. An hour of editing cannot beat ten minutes of reshooting then.
- Clipping through whole sentences. De-clip tools patch isolated peaks. When crackle runs through regular speech, the information is gone and repair is guesswork.
- The voice barely rises above the noise. If fan or traffic sits within about 10 dB of your voice, removing the noise removes chunks of voice with it, and the result sounds processed however careful you are.
- Wind across entire passages. A filter rescues gusts between phrases. Wind that swallows syllables has deleted them, and no tool invents missing words.
- Echo from a distant mic in a bare room. Mild reverb dries out. A camera mic three metres away in a tiled room captures more wall than voice, and de-reverb leaves a watery result.
- Someone else’s music under your speech. Songs from a cafe or passing car stereo cannot be separated cleanly from dialogue, and raise copyright problems on top of the clarity problem.
Everything above is a cure. Our guide to recording better vlog audio covers the prevention side: mic distance, recording levels and room choice, the decisions made in the minute before you press record that decide whether this repair chain is a quick polish or a lost cause.
Bad audio questions, answered
Short answers to the questions vloggers ask most about cleaning up audio in the edit. One rule runs through all of them: fix the specific fault, gently, in the right order.
Can background noise be removed from a video completely?
Rarely completely, and chasing complete removal damages the voice. Reduce steady noise until it stops drawing attention instead. A little room tone under clear speech sounds natural; total silence between words does not.
Why does my voice sound underwater after noise reduction?
That warbling sound is over-processing: the tool is removing speech frequencies with the noise. Halve the reduction, use two light passes, and check your profile holds no speech or breaths.
Should I remove noise before or after EQ?
Before. EQ boosts raise the noise along with the voice, giving noise reduction more to remove. The order is high-pass, noise reduction, EQ, de-reverb, limiter last.
What is a noise profile or noise print?
A short sample of the noise alone, a second or two with no speech in it, fed to the tool so it knows what to subtract. Capture it from a pause where only fan or hiss plays.
Can echo be removed from a video recording?
Partly. De-reverb tools subtract an estimate of the reflected voice, and a gentle gate trims tails in the pauses. Mild echo cleans up well; heavy echo from a distant mic never fully disappears.
How do I remove wind noise without making my voice thin?
Raise a high-pass filter slowly from near 70 Hz while a gusty passage loops, and stop when the roar drops away, usually at 80 to 120 Hz. Higher eats the body of the voice.
Can clipped audio be fixed?
Mild clipping can: de-clip tools rebuild the sliced-off peaks from the surrounding wave’s shape. Distortion through whole sentences cannot be rebuilt, because the detail was never recorded. Re-record that take at a lower level.
How loud should a YouTube video be?
About -14 LUFS integrated across the whole video, true peaks at or below -1 dBTP. YouTube normalises playback to roughly that level, so exporting at the target keeps you level with everyone else.
What is the best free way to remove background noise from a video?
Use your editor’s built-in tools first: most free editors include a denoise slider, EQ and a limiter. For stubborn steady noise, clean the exported audio in a free audio editor with a noise profile tool, then bring it back.
My recording is too quiet. Can I just turn it up?
You can, but the noise floor rises with the voice, so it gets louder without getting clearer. Denoise first at the recorded level, then raise level at the loudness stage, where the limiter manages peaks.
Keep exploring
Audio repair sits inside the wider edit. These guides cover the workflow around it.